welcome [tc18.tableau.com] · 2020-01-06 · ross bunker principal engineer tableau software...
TRANSCRIPT
Welcome
LOD Is the Key (to asking the right questions)
Ross Bunker
Principal Engineer
Tableau Software
# L O D I S T H E K E Y
Agenda
Why does LOD Matter?
What is LOD?
LOD in Tableau: A History
LOD in Tableau: The Future
Why does LOD matter?
If you do not know how to ask the right question, you discover nothing.
—W. Edwards Deming
If Tableau doesn’t let you ask the right question, you discover nothing.
—Me
What is the Question?
What are the total sales of each product? (Help me see by product type too)
What is the Question?
What is the salary of employees in each call center region?
The Replication Problem…
What is the salary of employees in each call center region?
What is LOD?
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Dimensions determine the granularity of your
measurements. They are the characteristics
of the objects you are measuring
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
#
#
#
#
Dimensions
Measures
Credit: Alan Eldridge, Sales Consultant Extraordinaire
What is LOD?
Many ‘things’ in a purchase order:Salesperson
Producer
Purchaser
Products
Company
State (salesperson)
State (purchaser)
State (producer)
Time (order date)
Time (ship date) not pictured
What is LOD?
Simple questions I might ask:
Total sales by date?
Unit of product X sold?
Highest volume
producer in each state?
Zip codes with
most purchases?
For this Given Set of Characteristics of These Things, What is My Measurement?
LOD is the Set of ‘Characteristics’ You Use to Define the ‘Things’ You are Measuring
LOD in Tableau:A History
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Tableau’s data model exposes a single flat
table as the LOD of the data source.
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
#
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Row LOD
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions in Viz
Viz LOD brings data from the ‘source’ LOD
to the LOD of my question. Viz is the
question and defines the LOD
Viz LOD
What are the total sales of each product? (Help me see by product type too)
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Features that let you summarize:
totals, ref lines, table calcs, etc.
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions
Summary LOD
Summary LOD
How do my individual product sales compare to the average sales for all products?
> 1000
SUM(Sales)
Customers
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Filters let you compute small
things at specific LOD.
(SUM(Sales) > 1000, Top N)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions
Filter LOD
How do sales of products fluctuate over time for products with lifetime sales > 40k?
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions in Viz
Extract at LOD let’s you redefine the
‘source’ LOD to reduce the amount of data
in your extract and improve performance
Extract Level of Detail
Extract LOD
What are my average monthly sales for each year/quarter?
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions in Viz
Geocoding lets you combine data based
on a subset of the LOD. LOD must be in
the viz. Data is replicated to VizLOD
State Latitude Longitude
WA 47.50 -120.50
OR 44.00 -120.5-
Geocoding
What is the sum of sales in each state and how does it vary by product type?
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Join LOD
Data Blending is a generalization of
geocoding. Arbitrary data aggregated to
a compatible LOD, joined, and replicated
(if necessary) to Viz LOD
#
#
#
Blending
How are my regional sales doing vs. quota?
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Join LOD
Data Blending was then extended to
support blending on dimensions not in
the viz. This introduced a ‘BlendLOD’
#
#
#
Blending at Join LOD
What is my profitability in each sales territory?
#
#
#
#
Totally Aggregated
Totally Disaggregated
(granularity of data source -
cannot go lower)
Gra
nu
larity
Less
More
Ag
gre
gatio
n
More
Less
#
#
#
Credit: Alan Eldridge, Sales Consultant Extraordinaire
Viz Level of Detail
Dimensions in Viz
Fixed LOD Expressions are
computed at a particular LOD,
then joined into the other results
at a compatible LOD
#
#
#
#
#
When used as a dimension, the
values are replicated at the data
source granularity
When used as a measure, the values
are joined at the LOD that causes the
least amount of replication and results
are then aggregated to VizLOD
LOD Calculations
Does the month in which a customer makes an initial purchase affect long term spend?
Asking Questions Requires Multiple Levels of Detail
There is no single view.
—Jock Mackinlay and Chris Stolte
There is no single LOD.
—Me
Questions that are Cumbersome from PO Data
What is my employee sales/salary ratio?Salary information is replicated to order LOD, need LOD calc to un-replicate
What employees have no sales in some months?Data is sales, so doesn’t show rows with employee and month if employee had no sale
Where are my customers relative to my stores?Plotting both on the same map today requires a join, but I want to show each independently
What is the average sales of my top ten customers in each state, compared to the rest of the customers in the state?
Ranking isn’t available in the database
LOD in Tableau:The Future
Future Improvements in Tableau
Richer computations even when not supported by the underlying database
Ability to combine data sources with different LODs into the same visualization without requiring a blend
Richer data modelling to support data sources that contain information at multiple levels of detail
Thank you!
#TC18
Ross Bunker
Principal Engineer
Please complete the
session survey from the My
Evaluations menu
in your TC18 app